Hi,
It appears that in the training dataset, T1W, T2W, and FLAIR images of the same subject have been co-registered and re-sampled, with a brain mask applied to crop these images.
Could you please clarify the image preprocessing steps applied to the training data?
Will the validation and test datasets be preprocessed in the same manner?
Thanks a lot.
Created by Ying Xia YX_aehrc Hi, please checkout the stable version v3, it should be updated now Thank you for updating the images. I've re-downloaded and checked the new data. Only the FLAIR image for HD265 has not been updated yet. Hi, Thank you for your feedback, this is really important! I have updated the flair due to the co-registration issue. Thanks. There are N = 10 training datasets whose FLAIR images were mis-aligned with T1W and T2W.
Here is the list: HD022, HD025, HD033, HD072, HD134, HD196, HD214, HD227, HD245, HD265.
Is it possible to fix their co-registration from your side given that we only have the cropped brain images?
yes, your understanding is correct, all modalities (T1/T2/FLAIR) are co-registered and resampled to the same resolution, and yes you may assume the validation and test datasets are pre-processed in the same way. we will update a more detailed pre-processing pipeline soon.